Temporality Based Sentiment Analysis Using Linguistic Rules and Meta-Data

Abstract

Internet is frequently used as a medium for exchange of information and opinion. The raw data available over the web is refined for the use in automated decision support system. Present day sentiment analysis generates Sentiscore by giving equal weightage to all the reviews without considering the temporal aspect. This somehow degrades the reliability of decision support system. In this paper, temporal sentiment analysis is proposed based on meta-data in conjunction with the linguistic context of words. An algorithm is also developed for evaluating Tempo-Sentiscore, which is a numeric value used to capture the temporal sentiment analysis. The proposed algorithm is evaluated on benchmark product review data. The experimental results based on temporality show high performance levels with precision, recall over 90%. The performance of the system demonstrates the effectiveness of the technique using human annotation. This paper also shows how the star rating is affected when Tempo-Sentiscore is considered in place of Sentiscore. This new star rating is close to the real scenario, i.e. Human Annotation.